4. Summary and discussion
Internal wave signatures are apparent in variations of descent rate data (w’ ) from Deep SOLO floats. Mean deep w’ variances (Figure 4) and dominant vertical wavelengths (Figure 5) exhibit geographical variability tied to proximity to rough seafloor topography and deep currents, with variations among and even within some deep basins. For example, within the Brazil Basin, mean deep w’ variances in the western part of the basin, where bottom roughness is very low on the abyssal plain, are on average about 3 times lower than those in the eastern part of the basin, where topography is very rough on the western flank of the Mid-Atlantic Ridge. This result is consistent with earlier findings of increased microscale energy and mixing over mid-ocean ridges (Polzin et al., 1997). Another interesting contrast in mean deepw’ variances is the relatively high values found within the Samoan Passage compared to the lower values immediately to the south of that passage. This elevated activity could be partly owing to the influence of lee waves and hydraulic jumps as bottom water flows northward through the passage (Carter et al., 2019). The elevated mean deep w’ variances further north of the passage could also be partly owing to the interactions of tidal flows, which can propagate over long distances, especially those at low vertical modes (Zhao et al., 2016), with local rough bathymetry.
Overall, mean deep w’ variances are correlated at 0.16 with local topographic roughness. So, as expected, internal wave activity is higher in the vicinity of rough topography. While this correlation is not extremely high, it is easily statistically significant (p=0.00). Furthermore, the dominant vertical wavelengths of w’ are also correlated at -0.07 with local topographic roughness (again, with p=0.00). This statistically significant negative correlation is also expected as locally generated internal waves should have shorter vertical wavelengths than those propagating from more distant generation regions, because the shorter vertical wavelength packets are more subject to breaking and subsequent dissipation.
It would also be interesting to examine internal wave signatures in deep density vertical strain using the float scientific data (temperature, salinity, and pressure profiles), similarly to what has been done with core Argo data (Whalen et al., 2012). However, many of these Deep SOLO floats were set to sample at 50 dbar intervals over some of the deeper portions of the water column to conserve float battery energy, and sampling at more frequent intervals field (25 dbar would be sufficient) would be required to properly estimate the deep strain. Nonetheless, the data we have so far of deep w’ variances provides a tantalizing glimpse of what might be learned of internal wave energy distributions in the deep ocean from a global Deep Argo array. Already, with only a few Deep Argo regional pilot arrays established, there are roughly three times the number of Deep Argo profiles than were used from all of WOCE in one study of deep mixing (Kunze et al., 2006).
Since float profiling rates typically take tens of seconds to adjust to changes in forcing including ambient vertical velocity, float buoyancy, and ambient density (e.g., Cusack et al., 2017), the perturbations in their descent rates will be slightly attenuated and lagged relative to the actual internal wave vertical velocities. However, even for the shortest dominant vertical wavelengths (e.g. 393 dbar) and the fastest background float descent rates (e.g. 0.17 dbar s-1), profiling for half a vertical wavelength would take around 20 minutes, far longer than float adjustment times. Hence, these measurement errors should have a very small impact on the results of this study.
For internal waves generated at the sea-floor, which would have group velocity (and energy) propagation upward, hence phase velocity (crest and troughs) propagation downward, sampling on descent would bias vertical wavelength estimates towards long values. For such internal waves at the inertial period, the dominant vertical wavelengths estimated here would be biased long by 10 (±6) %. If instead the energy propagation were always downward, the wavelength estimates would be biased short by 8 (±4) % at inertial periods. For internal waves with periods shorter than inertial, these errors would be larger. However, if energy propagation were evenly split between upward and downward packets, these errors would instead be random, and unbiased. At any rate, wave amplitudes should be unaffected by this issue.